Javanese adversatives, passives, and Mapping Theory

Javanese Adversatives, Passives and Mapping TheoryAuthor(s): William D. DaviesSource: Journal of Linguistics, Vol. 31, No. 1 (Mar., 1995), pp. 15-51Published by: Cambridge University PressStable URL: http://www.jstor.org/stable/4176298 .

Accessed: 08/11/2014 20:02

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

Cambridge University Press is collaborating with JSTOR to digitize, preserve and extend access to Journal ofLinguistics.

http://www.jstor.org

This content downloaded from 128.255.6.125 on Sat, 8 Nov 2014 20:03:00 PMAll use subject to JSTOR Terms and Conditions

http://www.jstor.org/action/showPublisher?publisherCode=cup

http://www.jstor.org/stable/4176298?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp


J. Linguistics 3I (I995), I5-51. Copyright © 1995 Cambridge University Press

Javanese adversatives, passives and Mapping Theory1

WILLIAM D. DAVIES

Department of Linguistics, University of Iowa

(Received 8 March I994; revised 3 August 1994)

Relational Grammarians have proposed Union analyses for adversatives many languages. An odd asymmetry in base predicates contrasts Japanese (Dubinsky 1985), in which adversatives may not be formed on unaccusatives, and Indonesian (Kana I986), in which adversatives may only be formed on unaccusatives. A close examination of adversatives in Javanese (a language closely related to Indonesian) resolves this asymmetry, revealing that Javanese adversatives (and by implication Indonesian adversatives) are best analyzed not as Unions but as passives. However, the passive analysis violates Perlmutter & Postal's (I984) i-Advancement Ex- clusiveness Law, which figures crucially in Dubinsky's elegant account of the distribution of Japanese adversatives. Gerdts' (I993a) Mapping Theory, rooted in an RG tradition, provides a solution, in which it is possible to capture the similarities of adversatives and other Javanese passives and at the same time preserve the insights of Dubinsky's analysis of Japanese.

I. INTRODUCTION

Constructions referred to as 'adversatives' or 'adversative passives' are a fairly common feature of Asian languages. They have been identified in Chinese, Japanese, Korean, Austronesian languages, and others. In many Austronesian languages these constructions are identified by the circumfix ke- -an added to the root form of the predicate. This is true in Javanese, a Western Austronesian language spoken primarily on Java, the most populous island of Indonesia, as illustrated in (i).

(i) Montor-e Amir ke-tiba-n watu. car-DEF A AD-fall rock 'Amir's car got fallen on by a rock.'

[i] The Javanese reported on here is the familiar (or Ngoko) style spoken by young adults in east Java, particularly in Surabaya. Some refer to it as 'Surabayan'. Thanks go to my principal Javanese consultants Susanto Teng, Surtjahjo, Ika Damayanti and Djoko Luknanto. Thanks also to Chris Culy, Stan Dubinsky and Donna Gerdts for discussion of issues discussed here and comments on earlier versions of this work and to two anonymous referees for valuable comments and suggestions. Errors and omissions are solely the responsibility of the author. This work was supported in part by funds from the Graduate College and the Center for Asian and Pacific Studies of the University of Iowa.

15



WILLIAM D. DAVIES

The notion of adversity at play in (I) is the fact that Amir is adversely affected by the rock falling on his car. Although the subject of some discussion, the adversative in Javanese has not received a syntactic analysis much beyond the description of a clause in which the main predicate takes ke- -an morphology.

Much, however, has been written regarding the adversative in Indonesian (Junus I97I, Cartier I978, Verhaar I984). In a recent Relational Grammar (RG) treatment, Kana (I986) adopts Dubinsky's (1985) clause union analysis of Japanese adversatives and modifies it to fit the Indonesian facts. Under the union analysis the predicate with adversative morphology is a complex predicate consisting of the base predicate and ke- -an, an abstract affective predicate. In Japanese the affective predicate introduces an argument or, minimally, assigns an 'affectee' thematic role to one of the arguments of the base predicate. Details of the union analysis are given in section 2.

In this article I argue explicitly against such an analysis for Javanese. Rather I demonstrate that so-called Javanese adversatives such as (I) are best analyzed as regular passives with separate morphology rather than as unions. However, the proposed analysis, first carried out within the RG framework, violates the I-Advancement Exclusiveness Law (Perlmutter & Postal I984). I demonstrate the superiority of an alternative analysis that captures the similarities of adversatives and other Javanese passives within Gerdts' (I992, I993a,b) Mapping Theory extension of RG. I thus propose that Javanese passives provide support for Gerdts' nascent theory.

The paper is organized as follows. Section 2 presents the union analysis of adversatives. In section 3 I illustrate the range of Javanese passive constructions, arguing for the appropriateness of a passive analysis. In section 4 I1 present evidence for the passive analysis of Javanese adversatives. Section 5 presents the RG analysis for Javanese and with it the problem for the i-Advancement Exclusiveness Law. In section 6 I outline the Mapping Theory solution, and in section 7 discuss residual issues.

2. THE UNION ANALYSIS OF ADVERSATIVES

Dubinsky (I985, I993) has proposed a clause union analysis for Japanese adversatives, such as that in (2).

(2) Tanaka ga sensei ni Hanako o sikarareta. T NOM teacher DAT H ACC scold.PASS.PERF

'Tanaka suffered the teacher scolding Hanako.' Lit. 'Tanaka was scolded Hanako by the teacher.'

Dubinsky motivates the union structure in (3) for (2).

I6



JAVANESE ADVERSATIVES, PASSIVES AND MAPPING

(3) ^;><-2 1 2 P

1 ^ 1 y

Cho Cho Cho

Cho Cho

Tanaka sensei Hanako sika rare

In (3), which uses the monoclausal union representation of Davies & Rosen I988, the 'inner clause' consists of the predicate sika 'scold' and the nominals Hanako and sensei 'teacher'. In the second P-sector,2 an affectee argument, here Tanaka, is introduced by the affective predicate rare and subsequently advances to I. Dubinsky's analysis accounts for a range of properties exhibited by Japanese adversatives.

Dubinsky's analysis has been applied to other languages as well, including Indonesian (Kana I986) and Korean (Gerdts I991). However, there is an asymmetry that sharply contrasts Japanese and Indonesian adversatives. Japanese permits adversatives of unergatives (4), but disallows adversatives with an unaccusative base (5).

(4) Tanaka ga kodomo ni ie de asobareta. T NOM child DAT house LOC play.PASS.PERF 'Tanaka suffered the child playing in the house.'

(5) * John ga kion ni agarareta. J NOM temperature DAT rise.PASS.PERF

(The temperature went up on John.)

Conversely, according to Kana (i986), Indonesian disallows adversatives of unergatives (6), essentially restricting adversatives to unaccusative bases (7).3

(6) * Sri kebicaraan pemuda. S AFFECT.speak young.man

(Sri endured a young man's speaking.)

(7) Mobil itu kejatuhan batu. car that AFFECT.fall rock 'The car got fallen on by a rock.'

An additional difference between Japanese and Indonesian is in the argument structure associated with adversatives. In Japanese, adversatives

[2] A P-SECTOR consists of all strata in which a particular predicate bears the P relation. Thus, in (3), the first stratum comprises the P-sector determined by the predicate sika 'scold' and the second and third strata comprise the P-sector determined by the predicate rare.

[3] Kana lists five transitive predicates that can occur in the Indonesian adversative: curi 'steal', copet 'pickpocket', lihat 'see', denger 'hear' and dapat 'find'. There is no speculation as to why these particular transitive predicates are available as bases of adversatives.

17

WILLIAM D. DAVIES

frequently have an argument structure distinct from active counterparts, the affective predicate rare introducing an additional argument. This is true in (2) above, where Tanaka can occur in the adversative. However, there is no active counterpart to (2), as (8) shows.

(8) * Sensei ga Tanaka o/ni Hanako o sikatta. teacher NOM T ACC/DAT H ACC scold.PERF

(The teacher scolded Tanaka Hanako.)

(8) is fully grammatical when Tanaka o/ni is omitted. Conversely, Indonesian adversatives rarely if ever introduce arguments. The subject of the adversative generally bears an oblique relation to the base predicate and can usually occur in an adpositional phrase in a nonadversative use of the base predicate. Compare (9) with (7).

(9) Batu itu jatuh di-atas mobil itu. rock that fall on-top car that 'The rock fell on the car.'

Dubinsky (I985, I993) accounts for the impossibility of adversatives of unaccusatives through appeal to the I-Advancement Exclusiveness Law (IAEX) (Perlmutter & Postal I984), which limits to one the number of advancees to I in a given clause. In the relational network (RN) in (io), the representation of (4), the adversative of an unergative, no universal laws or language-particular conditions are violated.

(Io) 2 1 L p

1--..~ 1 Loc Cho p , Cho Cho

Tanaka kodomo ie asob rare

However, in (i I), the representation of (5), the adversative of an unaccusative, contains a violation of the IAEX.

(II) * 1 2p

1 C 1 Choh\ Cho cho p

John kion aga rare

As an unaccusative, aga 'rise' determines an initial stratum containing a 2 but no I, here the nominal kion 'temperature'. In (II), both kion and John advance to I, thus violating the IAEX. Therefore, the clause represented by this structure is ill-formed.




Rather than appealing to any universal properties, Kana must resort to a language-particular, construction-specific stipulation in accounting for the Indonesian facts. She simply asserts that in Indonesian only unaccusative predicates can serve as the base for adversatives. Thus, on the face of it, it appears that the language-particular constraint of the Indonesian analysis undermines the appeal to universal principles to explain the Japanese case. From the standpoint of linguistic theory, this is an undesirable circumstance.

Examining data from Javanese can clarify the situation. As the data in (12)-(I3) show, the restriction on adversatives parallels that in Indonesian: unaccusatives appear to form adversatives (I2) whereas unergatives do not (I3).

(12) Montor-e Amir ke-tiba-n watu. car-DEF A AD-fall rock 'Amir's car was fallen on by a rock.'

(I3) *Kertas-e Amir ke-playo-nan bocahbocah. paper-DEF A AD-run children

(Amir's paper got run on by the children.) The Javanese adversative also parallels the Indonesian case in argument structure. The adversative subject generally bears an oblique relation to the based predicate and can occur in an adpositional phrase in a nonadversative counterpart, as in (I4).

(I4) Watu-ne tiba ning montor-e. rock-DEF fall on car-DEF 'The rock fell on the car.'

Despite parallels with Indonesian, a union analysis is inappropriate for the Javanese construction. Rather, the so-called adversatives are best analyzed as NON-VOLITIONAL accidental passives, and the restriction to LARGELY

unaccusative bases follows from an independently necessary language- particular constraint on advancement to 2. In section 4, I show that Javanese adversatives share properties with two other passive constructions: accidental passives and locative agentive passives. First, however, I examine some properties of Javanese passive and unaccusative constructions.

3. JAVANESE PASSIVES

One aspect of Javanese and closely related Western Austronesian languages that has received widespread attention is focus morphology, verbal morphology which indicates the roles of various arguments of the verb in a particular sentence. Analyses have identified the italicized morphology in the sentences in (I 5) in various ways.

(I 5) (a) Amir nempeling Bambang. A AF.slap B 'Amir slapped Bambang.'

I9



WILLIAM D. DAVIES

(b) Bambang di-tempeling (karo) Amir. B OF-slap by A 'Bambang was slapped by Amir./Amir slapped Bambang.'

The (homorganic) nasal prefix in (I 5a) has been called 'actor focus' (Naylor 1978, Thomas I980 for the Indonesian analogue), 'actor trigger' (Cumming I986, Wouk I986 for Indonesian and Malay), and 'active' (Bintoro I980, Ramelan I983, Poedjosoedarmo I986, Smith-Hefner I988, and others for Javanese). The di-prefix in ( 5b) has been labelled 'nonactor' or 'object focus' (Naylor 1978, Thomas I980, Smith-Hefner I988), 'patient trigger' (Cumming I986, Wouk 1986) and 'passive' (Bintoro 1980, Ramelan I983, Poedjosoedarmo I986, and countless others). The controversy lies in whether or not the notion 'subject' is assumed to play a role in Austronesian languages and the fact that discourse factors determine the appropriate morphology, thus making the Javanese passive functionally distinct from the passive of Indo-European languages.4 I will use the labels AF for actor focus and OF for object focus throughout this discussion.

I have argued elsewhere (Davies 1993) that subject is a necessary notion in the grammar of Javanese and that the di- construction is structurally a passive despite its different functional load in the grammar.5 Thus, the (RN) in (I6) is associated with the active sentence in (I 5a) and the passive RN in (I7) is rightly associated with the sentence in (I 5b).6

(I6) p 2

tempeling Amir Bambang

(I7) P I 2

Cho

tempeling Amir Bambang

[4] See Poedjosoedarmo (1986) for discussion of the discourse factors determining choice of active or passive morphology.

[5] This view of Javanese grammatical structure is in basic agreement with the analysis in Poedjosoedarmo (1986).

[6] The RNs in (i6)-(I7) are a slight simplification as I propose in Davies (I993) that it is also necessary to recognize a sentence-level TOPIC function. However, the additional complication of representing TOPICs in the diagrams will only obfuscate the points to be made here; nothing in the present discussion depends on recognition of the TOPIC function.

20




The di-prefixed verb represents one of a number of constructions that can be categorized as passive. The di-passive is the volitional or agentive passive, in which a theme argument, Bambang in (I5b), occurs preverbally and a third person agentive NP follows, optionally marked with the preposition karo 'with, by', Amir in (i5b).

The other passive of interest is the ACCIDENTAL or NONVOLITIONAL PASSIVE, illustrated in (i8).7

(I8) Bambang ke-tempeling Amir. B PASS-slap A 'Bambang got slapped by Amir.'

Here the prefix ke- marks the construction, which generally indicates that the agent performed the action accidentally. The RN associated with (I8) is basically the same as that for (i5b), the only differences being in fine-grained detail of morphological marking.

Given the potentially controversial assertion regarding passive in Javanese,8 I wish to discuss data that support the passive analysis. However, in order to do so, it is necessary to first discuss unaccusatives in Javanese. Additionally, the motivation of the unaccusative analysis for certain Javanese clauses interacts crucially with the problem facing the IAEX discussed in section 5.

3.1 Javanese unaccusatives

There is a morphological split among Javanese intransitive verbs. While some intransitives occur with an 'active' prefix, others do not; (I9) provides a partial list of verbs taking the prefix, and (20) is a list of verbs that do not.

(I9) mlaku 'walk' (20) tiba 'fall' mlayu 'run' teka 'arrive' nangis 'cry' lungguh ' sit' nglangi 'swim' pindah 'move'

[7] There is yet another 'passive' construction, which will not be discussed in detail here, in which the verb kenek 'hit' takes the place of preverbal morphology, as in (i).

(i) Bambang kenek tempeling Amir. B hit slap A 'Bambang got slapped by Amir.'

This sentence type usually carries adversative connotation. The productivity of the kenek +V structure varies from speaker to speaker and appears to be most prevalent in speakers from Surabaya.

[8] A number of researchers (e.g. Cumming I986, Durie I986, Nichols I990) have explicitly rejected passive analyses for Austronesian languages, but it seems largely because the construction has different discourse functions than passive in European languages.

21



WILLIAM D. DAVIES

mlumpat 'jump' cuwil 'chip' nguyuh ' urinate' rubuh 'collapse' mlebu 'enter' ilang 'disappear' mainan 'play' kerem 'run aground' ngomong 'say' klelep 'drown' nari ' dance' pecah 'break' njoked 'dance' kuatir 'worry' ngakon 'tell' semapot 'faint'

The presence or absence of the nasal 'active' morphology is not a perfect indicator of some notion of active vs. inactive. There are in fact a number of predicates that one might suppose to be in the control of the agent of the action that do not take the relevant morphology and some predicates that one might suppose to be outside the control of subject of the action that appear to take the nasal prefix. For example,

(2I) kandha 'say' (22) mbledos 'explode' takon 'ask' mbenedhol 'swell'

Therefore, although morphology provides a general key to the notions active and inactive, there appear to be verbs that can cross-over, and some that can actually occur with or without the prefix, e.g. omong' say'. However, roughly speaking prefixed intransitives seem to fall into the category of unergatives and unprefixed intransitives seem to fall into the category of unaccusatives.9

There are two syntactic environments that divide the intransitives somewhat more neatly into unergative and unaccusative and thus constitute crucial evidence for an unaccusative analysis for Javanese. The first is related to the periphrastic causative construction in which a clause is embedded under a verb of causation such as marahi 'make', ngongkon 'order', nggawe 'make', and others. The basic form of the construction is illustrated in (23)- (25).

(23) Bocah kuwi marahi bapak-e tiba. child that AF.make father-DEF fall 'The child made his father fall.'

(24) Bapake-e ngongkon bocahbocah nglangi ning kali. father-DEF AF.order children AF.swim in river 'Father made the children swim in the river.'

(25) Bocah kuwi marahi Amir nempeling Bambang. child that AF.make A AF.slap B 'The child made Amir slap Bambang.'

[9] See Kana (I986) for a discussion of the unergative/unaccusative split and ber-prefixation in Indonesian.

22




With certain intransitives, the order of the embedded subject and verb can be inverted. (26)-(27) illustrate.

(26) Bapak-e ngongkon nglangi bocahbocah ning kali. father-DEF AF.order AF.swim children in river 'Father made the children swim in the river.'

(27) Aku sing marahi nangis bayi-ne. I REL AF.make AF.cry baby-DEF 'I am the one who made the baby cry.'

Note that in (26), unlike (24), the subject of the embedded clause bocahbocah 'children' follows the embedded verb nglangi 'swim'. In the same way, bayi- ne 'the baby', the embedded subject in (27), follows nangis 'cry', the embedded verb. Certain other intransitives do not allow inversion of subject and verb, (28)-(29).

(28) * Bocah kuwi marahi tiba bapak-e. child that AF.make fall father-DEF

(The child made his father fall.)

(29) * Bapak-e ngongkon mektu bocahbocah saka omah. father-DEF AF.order go.out children from house

(Father made the children go outside.) This difference can be viewed as diagnostic of an unergative/unaccusative split in Javanese. Inasmuch as embedded transitive subjects cannot invert with their verb in causatives, the constraint on subject-verb inversion in causatives should be stated as:

(30) In periphrastic causatives, embedded subject and verb can invert iff the predicate is unergative.

Additional support for the unergative/unaccusative split proves important to the establishment of the passive analysis. This is a possession construction in which the possessor is topicalized and the possessed NP extraposed.1° Third person possession is indicated in Javanese by affixing the definite suffix-e optionally followed by the possessor NP; if there is no following NP, the reference of the possessor is specified in the discourse or the NP is simply definite. (3I) illustrates.

(3I1) Ibu-ne Siti mangan iwak mother-DEF S AF.eat fish 'Siti's mother ate fish.'

[IO] Verhaar (1978) discusses what appears to be a similar (albeit more restricted) analogue in Indonesian under what he calls the 'additional argument of inalienable possession'. Although inalienable possession plays no role in the Javanese construction for the speakers I have consulted, the limited data cited by Verhaar point to a syntactic condition similar to that developed here for Javanese.

23



WILLIAM D. DAVIES

Under limited circumstances, the possessor may be topicalized and the possessed NP extraposed, as illustrated in the pairs of sentences in (32)- (33).11

(32) (a) Ibu-ne Amir teka. mother-DEF A come 'Amir's mother arrived.'

(b) Amir // teka ibu-ne. A come mother-DEF 'Amir's mother arrived.'

(33) (a) Ban-e montor-e Amir mbledos. tire-DEF car-DEF A explode 'Amir's car's tire burst.'

(b) Amir // mbledos ban-e montor-e. A explode tire-DEF car-DEF 'Amir's car's tire burst.'

We can refer to the construction in (32b) and (33b) as the POSSESSOR TOPIC construction. In many cases, this structure is impossible (34)-(37).

(34) (a) Anak-e Siti mlayu. child-DEF S run 'Siti's child ran.'

(b) * Siti // mlayu anak-e. S run child-DEF

(Siti's child ran.)

(35) (a) Anak-e Bambang turu. child-DEF B sleep 'Bambang's child is asleep.'

[II] In the data '//' signifies a pause. The fact that (32b) and (33b) have a possessor topic structure and not a possessor ascension structure is evident from interaction with raising predicates. Under a possessor ascension analysis, the possessor would be the final subject of the clause and thus should be able to raise in appropriate environments. While (ib) shows that it is possible to raise the host NP embedded under the object focus form of kira 'think', the topicalized possessor may not raise, as in (ic).

(i) (a) Bambang ngira ibu-ne Amir teka. B AF.think mother-DEF A come 'Bambang thought Amir's mother arrived.'

(b) Ibu-ne Amir di-kira teka. mother-DEF A OF-think come 'Amir's mother was thought to have arrived.'

(c) * Amir di-kira teka ibu-ne. A OF-think come mother-DEF

(Amir's mother was thought to have arrived.)

In fact, the possessor in the possessor topic construction exhibits the properties ascribed to topics in Davies (1993), but fails to exhibit subject properties.

24




(b) * Bambang // turu anak-e. B sleep child-DEF

(Bambang's child is asleep.)

(36) (a) Anak-e Amir nyolong radio. child-DEF A AF.steal radio 'Amir's child stole a radio.'

(b) * Amir // nyolong radio anak-e. A AF.steal radio child-DEF

(Amir's child stole a radio.)

(37) (a) Asu kuwi nyokottangan-e Amir. dog that AF.bite hand-DEF A 'The dog bit Amir's hand.'

(b) * Amir // asu kuwi nyokottangan-e. A dog that AF.bite hand-DEF

(The dog bit Amir's hand.) (c) * Asu kuwi nyokot Amir tangan-e

dog that AF.bite A hand-DEF (The dog bit Amir's hand.)

Considering for a moment only the intransitive clauses in (32)-(35), we see another syntactic manifestation of the split in intransitive predicates, one that identifies the same classes of predicates identified by the periphrastic causative construction. Those predicates which are closely aligned with predicates that have been identified as unaccusatives in other languages allow the possessor topic construction, while those predicates aligned with unergative predicates in other languages do not. (36) shows that transitive subjects do not participate in the possessor topic construction and (37) shows that the construction is not possible with objects. Thus, in the data above, unaccusatives permit this structure. Syntactically, unaccusative subjects are distinguished from unergative subjects and transitive subjects and objects in that they head both a 2-arc and a I-arc. Thus, I would suggest the following condition on the possessor topic construction.

(38) A possessor can be topicalized and its possessed NP extraposed iff the host NP (the possessor and possessed together) heads both a I- arc and a 2-arc.

The crucial condition in (38) is that the host head both a I-arc and a 2-arc. This condition is met by the final subject of a passive clause. Thus, a prediction of the passive analysis for Javanese is that subjects of di-clauses and ke-clauses should allow the possessor topic construction. As (39b) and (4ob) show, this prediction is correct.

(39) (a) Garisan-e Amir di-tugel (karo) Musa. ruler-DEF A OF-break by M 'Amir's ruler was broken by Musa.'

25



WILLIAM D. DAVIES

(b) Amir // di-tugel (karo) Musa garisan-e. A OF-break by M ruler-DEF 'Amir's ruler was broken by Musa.'

(40) (a) Kopi-ne Siti ke-ombe Ali. coffee-DEF S PASS-drink A 'Siti's coffee got drunk by Ali.'

(b) Siti // ke-ombe Ali kopi-ne. S PASS-drink A coffee-DEF 'Siti's coffee got drunk by Ali.'

Granting the correctness of the condition on the possessor topic-extraposed possessed NP construction, the data in (39)-(40) provide evidence for the passive structure.

It is important to note that the adversative also takes the possessor topic construction. In (4I), the possessor, Amir, is a topic, and the possessed NP, montor-e '(his) car' is extraposed.

(4I) Amir // ke-tiba-n watu montor-e. A AD-fall rock car-DEF 'Amir's car got fallen on by a rock.'

3.2 Additional evidence for the passive analysis

Two additional brief arguments for the correctness of the passive analysis of the agentive and accidental constructions come from subject-oriented properties. First, in Javanese, as in most languages of the world, only subjects of embedded clauses may raise to an argument position in the matrix clause. A small set of Javanese predicates allow raising when the verb takes passive morphology. One such verb is kira 'think'. Raising is illustrated in (42b).

(42) (a) Musa ngira (nek) Amir ng-antem Ali. M AF.think COMP A AF-punch A 'Musa thought that Amir punched Ali.'

(b) Amir di-kira Musa ng-antem Ali. A OF-think M AF-punch A 'Amir was thought by Musa to have punched Ali.'

In (42b), Amir the agent of the embedded clause, occurs as the subject of the higher verb kira 'think'.12 If the passive analysis is correct, the preverbal NP in the di- and ke-clauses should be able to raise when embedded under a predicate such as kira. As (43) and (44) show, this is correct.

[12] See Davies (1993) for evidence for raising based on pleonastic subjects.

26




(43) Ali di-kira Musa di-antem Amir. A OF-think M OF-punch A 'Ali was thought by Musa to have been punched by Amir.'

(44) Bambang di-kira Musa ke-tempeling Amir. B OF-think M PASS-slap A 'Bambang was thought by Musa to have been slapped by Amir.'

Additionally, the preverbal NP in adversatives can be raised, as in (44), showing that it bears the subject relation.

(45) Montor-e Amir di-kira ke-tiba-n watu. car-DEF A OF-think AD-fall rock 'Amir's car was thought to have gotten fallen on by a rock.'

A further diagnostic of the passive structure is control into adverbial clauses. Adverbial clauses introduced by subordinators such as polae/merga 'because', waktune 'when', supaya 'so that', and others may have a null element in preverbal position. This is illustrated in (46).

(46) Amir mlayu polae kasep. A run because late 'Amiri ran because hei was late.'

In (46), the reference of the missing element is determined by Amir. While this type of control is subject to semantic and pragmatic effects, it is clear that in active and passive clauses the preverbal NP is the preferred controller, although direct objects are also possible controllers.13 Consider the following:

(47) (a) Asu-ne mangan iwak-e polae luwe. dog-DEF AF.eat fish-DEF because hungry 'The dogi ate the fishj because iti/,j was hungry.'

(b) * Iwak-e di-mangan asu-ne polae luwe. fish-DEF OF-eat dog-DEF because hungry

'The fishj was eaten by the dog, because itj,*i was hungry.'

In (47a) asu-ne 'the dog' controls the reference of the null pronoun of the adverbial clause. The object, iwak-e 'the fish', is for some speakers also a possible controller; however, in this instance it is pragmatically implausible, and such an interpretation is clearly not readily available. Importantly, in (47b), despite the pragmatic plausibility of the agent of the di-clause as controller, such an interpretation is not available. The unavailability of such

[13] Object control is more felicitous in an example such as (i):

(i) Amir nyepak Ali polae nakal. A AF.kick A because naughty 'Amiri kicked Alij because hei/ was naughty.'

27



WILLIAM D. DAVIES

a reading cannot be attributed to postverbal position (see fn. 13). However, these facts can be explained in Javanese by restricting chomeurs (such as the agent of the passive) from controlling null pronouns in adverbial clauses.'4 Thus, the passive analysis of the di-clauses provides an account of the control phenomena.

The postverbal NP of the accidental construction shows the same inability to control into adverbial clauses.

(48) Bambang ke-tempeling Ali polae semberana. B PAss-slap A because careless 'Bambangi got slapped by Alij because hei/,j was careless.'

In (48) only the preverbal NP, Bambang, can control the reference of the null pronoun, the postverbal NP, Ali, cannot. Again, the same set of facts pertains to adversatives.

(49) Amir ke-pate-nan anak waktune ning Amerika. A AD-die child when in A 'Amiri had a childj die on him when hei*j was in the U.S.'

The data considered here provide evidence for the viability of a structural passive analysis of the di-clauses and ke-clauses. Similarities between adversatives and the passive constructions with respect to raising and control have also been established.

4. ADVERSATIVES AS PASSIVES

We now turn to further illustrating the correctness of a passive analysis for the Javanese adversatives. Specifically, the ke- -an construction is best analyzed as an accidental passive in which the final subject is thematically a goal, locative, or some other oblique.

A number of shared properties of adversatives and accidental and agentive passives are noted above. The following sections present further similarities between adversatives and both types of passives.

4. I Similarities between accidental passives and adversatives

Adversatives share a number of properties with accidental passives aside from the prefix ke- and those properties outlined above.

First, in an adversative either the action of the 'agent' is not volitional or the situation is unexpected (hence the notion in many cases of adversity); this is precisely the same situation as the accidental passive. The lack of volition of the agent in these constructions is exhibited in a number of ways. First,

[I4] This restriction on controllers is a language-specific restriction holding for Javanese and is not intended to restrict controllers universally.

28




unlike the agentive passive, the agent of the accidental passive and the adversative cannot be marked by karo.15

(50) Bambang ke-tempeling (* karo) Amir. B PASS-slap by A 'Bambang got slapped by Amir.'

(5I) Amir ke-teka-nan (* karo) guru-ne. A AD-visit by teacher-DEF 'Amir got visited by his teacher.'

Second, unlike di-passives, accidental passives and adversatives cannot take agent-oriented manner adverbs such as cepetcepet 'quickly' and atiati 'carefully'.

(52) Bambang di-tempeling Amir cepetcepet. B OF-slap A fast.fast 'Bambang was slapped by Amir quickly.'

(53) * Bambang ke-tempeling Amir cepetcepet. B PASS-slap A fast.fast

(Bambang got slapped by Amir quickly.)

(54) * Amir ke-kancing-an lawang cepetcepet/atiati A AD-lock door fast.fast/carefully

(Amir got locked out quickly/carefully.)

Finally, adversatives and accidental passives cannot be formed on causative bases, which denote a conspicuous agentivity. Morphological causatives are formed with the suffix -kne, -kna, or -kake, depending on the dialect of the speaker. Thus, the adversative in (55b) and the accidental passive in (56b) are ungrammatical, whereas the di-passive counterparts (55c) and (56c) are fully grammatical.16

(55) (a) Amir nyrita-kne crita menyang Ali. A AF.story-CAUS story to A 'Amir told the story to Ali.'

(b) * Ali ke-crita-an crita Amir. A AD-story story A

(Ali got told the story by Amir.) (c) Ali di-crita-ni crita Amir.

A OF-story-LOC story A 'Ali was told the story by Amir.'

[I5] Some speakers (but not all) will accept sentences in which the agent is marked by the preposition dening 'with'. Poedjosoedarmo (I986) also reports this.

[I6] Javanese predicates can take only one suffix. Therefore, in (55b) the causative morpheme is replaced by -an. However, the absence of the causative morpheme is not responsible for the ungrammaticality of (55b). In (55c), the causative ending is replaced by the locative suffix -ni without loss of grammaticality or causative meaning.

29



WILLIAM D. DAVIES

(56) (a) Bambang nuro-kne bayi-ne. B AF.sleep-CAUS baby-DEF 'Bambang put his baby to sleep.'

(b) * Bayi-ne ke-turo-kne Bambang. baby-DEEF PASS-sleep-cAus B

(The baby was put to sleep by Bambang.) (c) Bayi-ne di-turo-kne Bambang.

baby-DEF OF-sleep-cAus B 'The baby was put to sleep by Bambang.'

A second property distinguishing di-passives from ke-passives concerns auxiliaries. Di-passives can occur with any of the full inventory of aspectual and modal auxiliaries, as seen in (57).

(57) Bambang arep/ wis/ lagi/ isa/ mari di-tempeling Amir. B will/ already/ PROG/ can/ PERF OF-slap A 'Bambang will be slapped by Amir.' 'Bambang has been slapped by Amir.' 'Bambang is being slapped by Amir.' 'Bambang can be slapped by Amir.' 'Bambang has been slapped by Amir.'

Accidental passives (58) and adversatives (59) can only occur with bisa/isa 'can' and mari, a perfective auxiliary.

(58) Bambangisa/ mari/* arep/* lagi/ * wis ke-tempelingAmir. B can/PERF/ will/ PROG/ already PAss-slap A 'Bambang can be/has been slapped by Amir.'

(59) Siti isa/ mari/ * arep/ * lagi/ * wis S can/ PERF/ will/ PROG/ already ke-kancing-an lawang-e. AD-lock door-DEEF 'Siti can get/has gotten the door locked on her.'

The data above show that adversative share certain crucial properties with accidental passives, which is a natural result of an analysis that treats them as accidental passives.

4.2 Adversatives and local di-passives

Adversatives also share some grammatical properties with locative di- passives, di-passives in which the final subject bears the goal, location, or source thematic role. The constructions are marked by the -i locative suffix, as in (6o)-(6I).17

[17] The locative suffix occurs as -i when preceded by a consonant and as -ni when preceded by a vowel. In the discussion that follows, I refer to the suffix as -i for convenience.

30




(60) Bambang di-kirim-i bungkusan (karo) Siti. B OF-send-LOC package by S 'Bambang was sent a package by Siti.'

(61) Amir di-teka-ni (karo) Bambang. A OF-visit-LOC by B 'Amir was visited by Bambang.'

The -i suffix also occurs on actives in which a goal, location, or source is foregrounded (at times in preference to a theme argument), as in the active counterpart of (60) in (62a). A standard RG treatment of this phenomenon would take -i as registering the advancement of a 3 or oblique to 2; (62a) would thus have the RN in (62b).18

(62) (a) Siti ngirim-i Bambang bungkusan. S AF.send-LOC B package 'Siti sent Bambang a package.'

(b) p 3

kirim Siti bungkusan Bambang

It has been noted in some grammars of Javanese that -an is the counterpart of-i in the accidental passive (e.g. Home 196I, Poedjosoedarmo I986). Here I substantiate this analysis by demonstrating similarities between adversatives and locative di-passives.

First, allowing some variation among speakers, the predicates serving as bases for the two constructions are for all intents and purposes the same. Included in the set are: uncal 'throw', tutup 'close', kirim 'send', sepak 'kick', ciprat 'splash', gawa 'carry', banjir 'flood', ilang 'disappear', and many others.

Second, the constructions show parallel postverbal word order char- acteristics. First, the agent and the theme of each construction may occur as a bare NP and may occur in either order. Thus, in the locative passive in (63), either ordering of the theme banyu panas 'hot water' and agent Bambang is acceptable.19

[I8] Chung (1976) and Kana (I986) have suggested this analysis for the locative -i suffix in Indonesian.

[I9] Speakers vary with respect to word order possibilities, some rejecting the bare NP agent in any but immediate postverbal position. Additionally, the bare theme NP must precede the agent when the agent occurs as object of the preposition karo. Although there is speaker variation, speakers are consistent across the two constructions.

3I



WILLIAM D. DAVIES

(63) Siti di-ciprat-i {Bambang banyu panas/ banyu panas Bambang}. S OF-splash-Loc B water hot/ water hot B 'Siti was splashed with hot water by Bambang.'

As (64) shows, the same is true of the 'agent' and theme of the adversative.

(64) Siti ke-ciprat-an {Bambang banyu panas/ banyu panas Bambang}. S AD-splash B water hot/ water hot B 'Siti got splashed with hot water by Bambang.'

Additionally, the possible placement of the emphatic element Iho (which has various effects on the meaning of a sentence) and maneh/meneh 'again, more' are similar in the two clause types. Here I illustrate only with Iho. With a locative di-passive the emphatic element Iho may precede the agent NP only when the agent takes prepositional marking. Thus, (65), in which Iho follows the unmarked agent anak-e 'her child', is fine, while (66), in which Iho precedes the agent, is only grammatical when the agent takes prepositional marking.

(65) Siti di-ciprat-i banyu panas anak-e lho! S OF-splash-LOC water hot child-DEF EMPH 'Siti was splashed with hot water by her child!'

(66) Siti di-ciprat-i banyu panas lho *(karo) anak-e! S OF-splash-LOC water hot EMPH by child-DEF 'Siti was splashed with hot water by her child!'

The adversatives in (67)-(68) show the same constellation of facts. However, the variant in which the 'agent' is preceded by Iho cannot be saved since the 'agent' cannot take prepositional marking, (68).

(67) Siti ke-ciprat-an banyu panas anak-e lho! S AD-splash water hot child-DEF EMPH 'Siti got splashed with hot water by her child!'

(68) * Siti ke-ciprat-an banyu panas lho anak-e! S AD-splash water hot EMPH child-DEF

(Siti got splashed with hot water by her child!)

These facts underscore the close relationship of adversatives and locative di-passives. The analysis is further strengthened by the fact that accidental passives cannot take the locative -i suffix. The sentences in (69)-(70) are ungrammatical solely because of the form of the suffix.

(69) * Siti ke-ciprat-i banyu panas anak-e. S PASS-splash-LOC water hot child-DEF

(Siti got splashed with hot water by her child.)

32




(70) * Bambang ke-kirim-i bungkusan. B PASS-send-LOC package

(Bambang was (accidentally) sent a package.)

The -i suffix is possible in the accidental passive construction taking the verb kenek 'hit' rather than the prefixed form (see note 7).

(7I) (a) Siti kenek ciprat-i banyu panas. S hit splash-Loc water hot 'Siti got splashed with hot water.'

(b) Bambang kenek kirim-i bungkusan. B hit send-Loc package 'Bambang got sent a package.'

The fact that the -i suffix is possible here indicates that it is not inherently incompatible with nonvolitional or accidental meaning.

5. AN RG ANALYSIS AND A PROBLEM

Having established that Javanese adversatives are accidental passives with a general passive structure, it remains to account for their distribution. As passives with non-theme final subjects, adversatives such as (64) will have the representation in (72).

(72) 1C2/ 3\

P Cho Cho1 / - - / \ \Cho

ciprat Bambang banya panas Siti

Under the interpretation that the locative suffix registers advancement to 2, Siti advances from 3 to 2 and then advances to I in the passive structure.

The problem for the theory of RG emerges when we consider the case of unaccusative bases. For example, within a standard RG account, (73a) would have the structure in (73b).

(73) (a) Montor-e Amir ke-tiba-n watu. car-DEF A PASS-fall-LOC rock 'Amir's car got fallen on by a rock.'

(b) PP- 2/ 32

tiba watu montor-e Amir

33



WILLIAM D. DAVIES

Tiba 'fall' is an unaccusative predicate (as shown in section 3. I), so the initial stratum in (73b) will contain a 2-arc but no I-arc. In order for there to be a transitive stratum for the passive structure, the initial 2 must advance to I and the goal, montor-e Amir 'Amir's car' must advance to 2, as indicated in the representation. Montor-e Amir can then advance to i via passive. The problem, of course, is that the structure in (73b) violates the IAEX, and thus (73a) should be ill-formed.20

An alternative that does not violate the IAEX is direct 3-I Advancement. Advancement to i of non-2s has been suggested for passives in a variety of languages (e.g. Kimenyi i980, Joseph 1982, Suzuki I984). In a 3-I advancement analysis, (73a) would have the structure in (74).

(74) P 2 3 p / 2 1

tiba watu montor-e Amir

Clearly (74) contains no iAEX violation, and so the 3-I Advancement analysis is compatible with all proposed RG laws.

However, (73b) and (74) differ in a crucial respect. In (73b), the initial 2, watu 'rock', is a final chomeur, while in (74) it is a final 2. The two analyses thus make different predictions about the status of the initial 2 in regard to grammatical properties distinguishing chomeurs from final 2s. One such property examined above is control into adverbial clauses. As shown in section 3.2, final 2s can control into adverbial clauses, (47a) and (i) in fn. 13, while chomeurs cannot, (47b) and (48). As (49), repeated below, shows, the initial 2, here anak 'child', of an adversative with an unaccusative base, here pate 'die', cannot control the null pronoun in an adverbial clause.

(49) Amir ke-pate-nan anak waktune ning Amerika. A AD-die child when in A 'Amiri had a childj die on him when hei,j was in the U.S.'

Thus, the analysis of adversatives which violates the IAEX makes the correct prediction with respect to the behavior of the initial 2 of an unaccusative. Therefore, this appears to be the most reasonable solution within RG.

6. A MAPPING THEORY SOLUTION

As I attempt to show in this section, Gerdts' (1992, I993a,b) Mapping Theory, derived in large part from insights from RG, provides a solution for Javanese.

[20] Others have presented evidence against the IAEX from passives and impersonal passives of unaccusatives (Nerbonne I982, Timberlake i982, Ozkarag6z i986, Gerdts i988b).

34




6.I Some Mapping Theory basics

In an attempt to explain the constellation of RG structures found in particular languages, Gerdts develops a theory of morphosyntactic argument structure based on many of the same assumptions important to 'classic' RG. Mapping Theory representations specify four types of information about a nominal: (i) its thematic role, (ii) its grammatical relation, (iii) its MAP (morphosyntactically-licensed argument position) if it has one, and (iv) its presentation, language-particular statements regarding word order, case, agreement, and so on. The MAPs in a language correspond to the 'direct arguments', or the syntactically active arguments. A language will generally have a maximum of 2 or 3 MAPs depending on whether most morphosyntactic processes (e.g. structural vs. inherent case, verb agreement, word order) apply to only subjects and direct objects or to subjects, direct objects, and indirect objects. The MAP structure for a simple transitive clause in English (a 2-MAP language) is given in (75b).

(75) (a) Kelly trimmed the hedges.

(b) thematic relations: agent theme

I I grammatical relations: 1 1 (initial GRs in RG)

MAPs: A B (final GRs in RG) I I

presentation: non. case acc. case

MAPs are ordered positions (represented as A, B, etc.) linked to morphological statements (e.g. nominative case licenses A, immediate postverbal position licenses B). In any given clause, the number of MAPs is based on three things: (i) the valence of the verb, (ii) MAP-building or -reducing morphology, and (iii) the MAP threshold set for the language.

As suggested by the example in (75b), each of the modules is ordered and aligned with the others, and each module is linked only with the module(s) adjacent to it. The linking of modules is governed by a number of principles which are well-formedness conditions akin to laws of RG and conditions on linking in other linking theories. Gerdts (I993a) proposes the universal principles in (76).

(76) Principles for Linking GRs and MAPs (Gerdts I993a) (a) SATURATION PRINCIPLE: Every MAP must be linked to a GR or

cancelled. (b) BIUNIQUENESS PRINCIPLE: Every MAP is linked to a single GR

(except multiattachment under coreference), and every GR is linked to at most one MAP.

(C) NO DELINKING PRINCIPLE: There are no delinkings.

35



WILLIAM D. DAVIES

Finally, UNMARKED ASSOCLATIONS of GRs to MAPs proceed in a vertical left-to-right fashion, as illustrated in (75b). MARKED ASSOCIATIONS involve non-vertical linkings or linking a nominal not present in the basic valence of the verb. These are generally accompanied by morphological conditions and marking and parallel some of the relation-changing structures of RG, e.g. passive, 3-2 advancement, and so on. For example, the passive clause in (77a) has the representation in (77b).

(77) (a) The hedge was trimmed by Kelly.

(b) 0-Rs: AG TH

GRs: 1 2

MAPs: A I

pres.: nom. case

Within Mapping Theory, the general passive representation is characterized as follows:

(78) Passive (general) (i) the highest GR is not linked to a MAP, and

(ii) a MAP is cancelled if possible (i.e., if more than one MAP is present in the structure).2'

This is the most general statement of passive, intended to capture what is common to all languages. Grammars of particular languages will include codicils that further specify the realization of passive for that language. Thus,

[2I] It is not always the case that there will be a MAP to cancel, as in impersonal passives of intransitives and some other structures. An impersonal passive of an intransitive may have a representation such as (i).

(i) 0-Rs: AG

GRs: 1 2

MAPs: A

In (i), nonthematic 2 (a 2 linked to no thematic argument) occurs in the structure and is linked to the A MAP. Here no cancellation of a MAP is possible. The nonthematic element is the er of Dutch impersonal passives, the es of German impersonal passives, and so on.

36


the precise MAP that is cancelled is specified in the grammar of the language. In English, as in many other languages, the B MAP is cancelled and must be obligatorily cancelled (ruling out the possibility of impersonal passives); the cancellation is denoted by the angle brackets in (77). However, for Halkomelem Salish, Gerdts (I993a) has proposed that the A MAP is cancelled. (77) instantiates the passive in English since the theme 2 links directly to the A MAP, the agent i goes unlinked, and the B MAP is cancelled.

I now briefly illustrate how two of the above linking principles work and discuss the relationship of Mapping Theory and Relational Grammar. According to the Saturation Principle (76a), (75) and (77b) are well-formed representations because each MAP is either linked or cancelled. The Saturation Principle would rule out a representation such as (79), however.

(79) 0-Rs: AG TH I I

GRs: 1 2

MAPs: A B I I

pres.: nom. case acc. case

(79) violates the Saturation Principle because the B MAP is not linked. Likewise, (75) and (77b) satisfy the Biuniqueness Principle (76b). Although in (77b) the I is not linked to a MAP, in neither structure is there a one-to- many mapping between GRs and MAPs. This illicit state of affairs is illustrated in the hypothetical structure in (8o), which is ruled out by the Biuniqueness Principle inasmuch as the I is linked to both the A and B MAPs.

(80) O-Rs: AG

GRs: 1

MAPs: A B I I

pres.: nom. case acc. case

These linking principles thus work as do the laws of RG, as well-formedness conditions on representations.

Although inspired by the insights of Relational Grammar and sharing many assumptions with that theory, Mapping Theory is a departure in a number of ways. First, Mapping Theory is a two-level theory. While RG can incorporate a number of distinct strata of GRs, Mapping Theory

37



WILLIAM D. DAVIES

incorporates only the levels of GRs and MAPs. This automatically reduces the number of possible structures recognized by the theory. For example, Gerdts (I993a) proposed that antipassives have the following structure.

(8I) 0-Rs: AG TH I I

GRs: 1 2

MAPs: A 

Rather than the I to 2 demotion proposed by Postal (1977), which accounts for the chomage of the (initial) 2 of antipassives, the theme 2 is simply not linked to a MAP and the B MAP is cancelled.22

This points to a second distinction between Mapping Theory and RG, the status of the chomeur relation. Rather than positing a separate chomeur relation, the notion of ch6mage is reconstructed as a term relation (I, 2, 3) that goes unlinked to a MAP. Thus, the agent of a passive is an unlinked I in Mapping Theory and a i-chomeur in RG, the theme of an antipassive is an unlinked 2 in Mapping Theory and a 2-chomeur in RG, and so on.

Third, Mapping Theory attempts to explain the clustering of structures proposed in RG treatments by identifying the number of MAPs that a language has. As Gerdts (I992) shows, a particular language tends to have advancements (and demotions) to 2 or to 3, but not both. Within Mapping Theory, this fact follows from whether or not the language has a 2- or 3- MAP threshold, while within RG, there is no current explanation for this clustering of structures.

Finally, Mapping Theory allows accounts of various language-particular facts to follow from general principles in a way unavailable in classic RG. To give one example, Gerdts (I993a) shows that the RN in (82) is ill-formed in Halkomelem Salish.

(82) P 1/ E

1 Cho

P 1/ Co Cho ,C

There is no currently available explanation within RG other than a language- particular constraint ruling out antipassives of non-initial 2s. By contrast, this

[22] Whether or not the data providing empirical evidence for the 2-hood of the subject of antipassives presented in Davies (I984) and Davies & Sam-Colop (I990) can be accounted for within Mapping Theory is left an open question here.

38

is ruled out in Mapping Theory because the corresponding representation in (83) is ruled out.

(83) 0-Rs: AG TH BEN

GRs: 1 2 Obl

MAPs: A 

(83) is ill-formed due to a violation of the No Delinking Principle. The applicative construction (oblique to 2 advancement in RG) requires the oblique to link to a MAP (the B MAP here). However, antipassive requires the B MAP to be cancelled. The only way to achieve both of these is to link and delink the Obl to B MAP linking, which violates the No Delinking Principle.

6.2 A Mapping Theory solution for Javanese

Mapping Theory provides a solution to the problem of passives of unaccusatives in Javanese. Gerdts (I993a) identifies Indonesian as a 2-MAP language, and Javanese is a 2-MAP language for many of the same reasons, i.e., strict subject-verb word order, advancement is to direct object, subjects and direct objects are the 'active' nominals. This means that a Mapping Theory representation of an active transitive clause would be as in (84).23

(84) (a) Amir nempeling Bambang. A AF.slap B 'Amir slapped Bambang.'

(b) O-Rs: AG TH I I

GRs: 1 2

MAPs: A B

Javanese passives parallel English passives, as illustrated in (84).

(85) (a) Bambang di-tempeling (karo) Amir. B OF-slap by A 'Bambang was slapped by Amir.'

[23] In what follows, I ignore the presentation level of the structures since the focus of interest is GR to MAP associations.

39

WILLIAM D. DAVIES

(b) 0-Rs: AG TH I I

GRs: 1 2

MAPs: A 

In the passive, the theme 2 links directly to the A MAP, the agent I is unlinked, and the B MAP is obligatorily cancelled. Passive morphology (here the object focus prefix di-) is triggered when the highest GR goes unlinked. The fact that the agent is unlinked accounts for its being morphosyntactically inert. As stated above, grammars of particular languages will contain codicils detailing how the general passive structure is instantiated in that language. The following is the Javanese passive codicil.

(86) Javanese Passive Codicil (i) The highest GR is not linked to a MAP.

(ii) The B MAP is obligatorily cancelled.

Advancement to 2 (in RG terms) represents the other marked association of interest in Javanese. The sentence in (87a) has the structure in (87b).

(87) (a) Siti ngirim-i Bambang bungkusan. S AF.send-LOC B.package 'Siti sent Bambang a package.'

(b) O-Rs: AG TH GO I I I

GRs: 1 2 3

MAPs: A B

Since Javanese is a 2-MAP language, the way in which the goal 3 (here Bambang) can be made active is by linking to a MAP. In (87b), the agent I links to the A MAP (accounting for its preverbal position and the actor focus morphology on the verb) and the goal 3 links to the B MAP (accounting for its immediate postverbal position). LOC morphology is triggered when a 3 is linked to a MAP. The theme 2 is unlinked and is morphosyntactically inert.

Returning to the adversatives and treating them now as accidental passives, the representation of a ditransitive adversative is given in (88).

(88) (a) Siti ke-ciprat-an banyu panas Bambang. S PASS-splash-LOC water hot B 'Siti got splashed with hot water by Bambang.'

40

(b) 0-Rs: AG TH GO I I I

GRs: 1 2 3

MAPs: A 

Linking the goal 3 with either MAP results in LOC morphology, here -an since it is an accidental passive, and linking a lower-ranking GR to the A MAP triggers passive morphology, here ke-. Thus, linking the 3 to the A MAP accounts for all the morphology here. Cancellation of the B MAP is part of the passive structure in Javanese (86ii). Therefore, in (88) both the agent I and the theme 2 go unlinked and are morphosyntactically inert. Note that a locative di-passive will have the same representation: the -i morpheme is the morphological reflex of linking the goal 3 to a MAP when an accidental passive is not involved, as above in (87b).

Since the representations for agentive and accidental passives are identical, a rule for ensuring the appropriate passive morphology is necessary. This can be stated informally as:

(89) Javanese Passive Morphology Select ke- as passive prefix for accidental passives; otherwise select di-.

The informal condition in (89) simply ensures that the appropriate passive prefix will be selected depending on the semantics of the situation being described. As we have seen above, the form of the passive morpheme determines the form of the locative morpheme. Therefore, the rule for locative morphology must include some reference to the accidental passive morpheme.

(90) Javanese Locative Morphology The locative morpheme is suffixed to the verb stem iff a 3 links to a MAP. If the verb is prefixed with ke- the locative ending is -an; otherwise it is -i.

The informal rule in (90) ensures that -an occurs only with ke-, ruling out the ill-formed sequence *di-... -an.

Accepting an analysis such as (88) allows a solution to the problematic case of adversatives of unaccusative bases that does not violate the IAEX or any of Gerdts' mapping principles. The passive with an unaccusative base is represented as in (9Ib).

(9I) (a) Montor-e Amir ke-tiba-n watu. car-DEF A PASS-fall-LOC rock 'Amir's car was fallen on by a rock.'

41

WILLIAM D. DAVIES

(b) 0-Rs: TH GO I I

GRs: 2 3

MAPs: A 

The representation in (9gb) results in the appropriate morphology. A 3 is linked to a MAP, hence the verb takes locative morphology. Additionally, the highest GR (here 2) is not linked to the A MAP, thus yielding passive morphology. Since this is an accidental passive, the form of the passive morpheme is ke- (following (89)) and the form of the locative suffix is -an (following (90o)).

Given the relationship of 'adversatives' and locative di-passives, one expects that locative di-passives with unaccusative bases should be possible. The correctness of this prediction is illustrated in (92).

(92) Bambang di-teka-ni Amir. B OF-come-LOC A 'Bambang was visited by Amir.'

The sentence in (92) will have the same representation as that given in (9Ib). However, since this is not an accidental passive, following (89), the passive morpheme is realized as di-, and following (90o), the locative morpheme is realized as -ni.

So it is clear that not only adversatives but also Javanese agentive passives provide evidence against the IAEX. The Mapping Theory treatment of these structures avoids this problem entirely.

7. CONCLUSION

There are a number of points that remain to be made and issues to be resolved. I turn to these now.

7. I Tertiary passives

Gerdts (I993a) outlines various points of Halkomelem Salish grammar for which a Mapping Theory account is superior to a Relational Grammar account. One of these is an account of passives of unaccusatives much like that presented above for Javanese. Others are more language-specific points. However, Mapping Theory also provides a solution to an old problem in RG regarding an unattested structure that should be possible.

Many languages contain both passives and advancements to 2. We have seen that above for Javanese. However, when passive and advancement to 2 are present in the same clause, it is invariably the advancee to 2 which is the final subject of the clause. That is, while we have seen instances of locative

42

passives, such as (93), there are no clauses with both locative and passive morphology such as (94), in which the theme is the subject.

(93) Amir di-kirim-i bungkusan-e (karo) Siti. A OF-send-LOC package-DEF by S 'Amir was sent the package by Siti.'

(94) * Bungkusan-e di-kirim-i Amir (karo) Siti. package-DEF OF-send-LOC A by S

(The package was sent Amir by Siti.) The RN for (94) would be:

(95)* p 1 / 32

Cho/ 1 ho

kirim Siti bungkusan-e Amir

In (95), the initial 2 advances to i in the second stratum and the initial 3 advances to 2 in the third stratum. The clause type represented in (95) has been termed a 'tertiary passive' by Postal (i986).24

No unassailable account for the absence of tertiary passives is available within RG.25 However, an explanation is readily available within Mapping Theory. As we have seen, locative passives such as (93) have the representation in (96).

(96) 0-Rs: AG TH GO I I I

GRs: 1 2 3

MAPs: A 

[24] As Bickford (I987) discusses at length, the status of tertiary passives in languages other than English is uncontroversial: they are ill-formed. Tertiary passives in English are subject to dialectal variation as well as lexical, syntactic and discourse constraints (see Fillmore I965, Green I974, Oehrle I976). An example is given in (i).

(i) # A book was given her by Pat.

Bickford provides an RG analysis of English tertiary passives in which the indirect object, here her, has not advanced to 2, but is simply a 3 without prepositional marking. In support of the analysis, Bickford cites the fact that prepositionless 3 analysis is no more complex than the analysis favored by Postal (I986) in which the 3 advances to 2 in a stratum subsequent to passive, the analysis associated with the structure in (95).

[25] Bickford (I987) proposes the Nuclear Novice Law to account for the impossibility of tertiary passives; however, robust counterevidence to this law is available from at least French (Legendre I990) and Korean (Gerdts I990). Therefore, I will not discuss this here.

43

WILLIAM D. DAVIES

The tertiary passive would have to have the Mapping Theory representation in (97).26

(97) 0-Rs: AG TH GO

GRs: 1 2 3

MAPs: A B The theme 2 is linked to the A MAP and the goal 3 is linked to the B MAP, accounting for syntactic and morphological properties one would expect in such clauses. However, (97) is not a well-formed representation. Recall that the Mapping Theory rule of passive for Javanese requires: (i) the highest GR goes unlinked and (ii) the B MAP is cancelled. Since there is a B MAP in (97), it must be cancelled for the structure to be a well-formed passive. Since it is not, the representation is ill-formed and the corresponding sentence will be ill-formed as well.

This account is not specific to Javanese. The passive rule, as stated in (78), requires the cancellation of a MAP if there is more than one MAP present. The tertiary passive requires the presence of more than one MAP; otherwise it is impossible for both the theme 2 and the goal 3 to be linked to MAPs. Therefore, in any 2-MAP language it is logically impossible to satisfy the requirements for both aspects of the tertiary passive structure. Mapping Theory thus provides a non-ad hoc, nonconstruction-specific account for the absence of tertiary passives.27

[26] One reviewer suggested that a Javanese tertiary passive could have the representation in (i), in which the B MAP is cancelled and the goal 3 is unlinked. (i) O-Rs: AG TH GO

I I I GRs: 1 2 3

MAPs: A 

While (i) is a well-formed representation, it is not a possible representation for (94), the Javanese tertiary passive. This is because the goal 3 remains unlinked; therefore, the condition for locative morphology is not satisfied. (i) is a well-formed representation for the sentence in (ii).

(ii) Bungkusan-e di-kirim (karo) Siti menyang Amir. package-DEF OF-send by S to A 'The package was sent to Amir by Siti.'

[27] The data discussed by Postal (1986) and Bickford (1987) make clear the fact that tertiary passives are not an issue in 3-MAP languages. Within Mapping Theory this follows from the generalization noted by Gerdts (I992, I993a) that the target of RG revaluations in 3- MAP languages is the 3 relation (the C MAP in Mapping Theory terms). This means that the 3-2 and Oblique-2 Advancement structures crucial to the tertiary passive configuration are excluded from the grammars of these languages.

44


7.2 The i-Advancement Exclusiveness Law

The solution to the violation of the IAEX inherent in the passive analysis of certain Javanese clauses has been to adopt a framework in which the IAEX is inoperative. The natural question to ask is what becomes of analyses dependent upon the IAEX. Can a Mapping Theory approach handle the data which need to be accounted for? Reanalyses of various constructions within RG have resulted in a great diminishing of the domain of data for which Perlmutter & Postal (I984) formulated the IAEX.

For example, Gerdts (i988a) proposes that Ilokano (and by extension other Philippine languages) are ergative and that the revaluation target is 2 rather than I. Thus, the lack of multiple advancements to I in Cebuano, which supplies some of Perlmutter & Postal's motivation for the IAEX, may in fact be irrelevant. With respect to another type of data cited by Perlmutter & Postal, in a recent paper, Richards (I994) has proposed that English pseudo-passives are best analyzed as unions. Within this analysis, Richards is able to account for the fact that only unergative (and not unaccusative) predicates occur in this construction without reference to the IAEX. Additionally, the fact that in some languages unaccusatives can occur in impersonal passive structures (as argued by Nerbonne I982, Timberlake I982, and Ozkarag6z I986) points to the fact that certain restrictions on passives should be relegated to language-particular statements and casts serious doubt on the relevance of these structures to establishing the IAEX.

However, some of the data presented by Perlmutter & Postal (I984) that remain can be simply accounted for within Mapping Theory. Among the data they cite are 'sporadic advancements to I' in English, as in (98b).

(98) (a) Melvin bought a lot of heroin for five dollars. (b) Five dollars bought a lot of heroin in I827.

Perlmutter & Postal propose that (98b) includes the sub-network in (99).

(99) p 2 GR

buy a lot of heroin five dollars

A Mapping Theory analogue would include the substructure in (100).

45



WILLIAM D. DAVIES

(ioo) 0-Rs: TH INSTR I I

GRs: 2 Obl

MAPs: A B

Perimutter & Postal go on to argue that despite the fact that the final stratum in (99) is a legitimate departure stratum for passive, the passive clause in (I o a) is ruled out since the associated RN ( o I b) contains two advancements to I.

(IOI) (a) *A lot of heroin was bought by five dollars in I827.

(b) P 2 /GRx

2 o

buy a lot of heroin five dollars

Given the No Delinking Principle (76c), illustrated above for Halkomelem antipassives, Mapping Theory easily accounts for the ungrammaticality of (Iooa). In addition to cancelling the B MAP, the structure would include the oblique five dollars linking to the A MAP and delinking so that the 2, a lot of heroin, can link to the A MAP. This is schematized in (102).

(I02) 0-Rs: TH INSTR I I

GRs: 2 Obl

MAPs: A 

Clearly, (I02) violates the No Delinking Principle, accounting for the ungrammaticality of (ioia). Thus, there is no well-formed representation of (Ioia) within Mapping Theory.

Perlmutter & Postal (1984) also cite the impossibility of the 'conceivable' passives in (Io3d,e).

(103) (a) Mark paid attention to Betty. (b) Betty was paid attention to by Mark. (c) Attention was paid to Betty by Mark. (d) * Betty was paid to by Mark by attention. (e) * Betty was been paid to by Mark by attention.

46


It is clear that multiple passives of this type or any other type will be ruled out within Mapping Theory for the same reason that (loia) is: any multiple passivization will require delinking.

The remaining case to consider is the one most relevant to this discussion, Dubinsky's appeal to the IAEX to account for the absence of adversatives of unaccusatives. I will demonstrate here that Mapping Theory can provide a satisfactory explanation.

Within Mapping Theory, Culy & Davies (1993) have proposed that union structures should be represented by what they term 'layered structures'. The idea is that when the union predicate or morpheme is added to a verb, its argument creates a new layer of GRs. The inner layer contains the GRs of the stem predicate, and the outer layer contains the GR linked to the added argument and the GR linked to the inner layer. This is illustrated by the Japanese causative in (104) (from Suzuki I984).

(104) (a) Tarooga imooto o kyoori ni kaeraseta. T NOM younger.sister ACC hometown to return.CAUS.PERF 'Taro made his younger sister go back to hometown.'

(b) O-Rs: Causer Agent Loc I I

GRs: [ 1 Obl

1 2

MAPs: A B

The inner layer of the layered structure in (Io4b) is enclosed in brackets. The agent I of the inner layer is linked to the B MAP by a combination of language-particular and universal linking conventions and the causer I is linked to the A MAP. Culy & Davies (1993) show the efficacy of layered structures for the analysis of Fula, a Niger-Congo language.

With respect to the Japanese affective construction, Culy & Davies propose that the sentence in (4), repeated here as (105), has the structure in (IO6).

(105) Tanaka ga kodomo ni ie de asobareta. T NOM child DAT house LOC play.PASS.PERF 'Tanaka suffered the child playing in the house.'

47



WILLIAM D. DAVIES

(I06) O-Rs: AG LOC Affectee I I

GRs: [1 Obl]

1 2

MAPs: A 

The layered structure enclosed in brackets in (106) represents the GRs of the stem predicate, asob 'play'. The affectee 2, here Tanaka, is introduced by the passive morpheme -rare, as in Dubinsky's analysis. The affectee is marked nominative as it is linked to the A MAP. The unlinked agent, kodomo 'child', takes dative marking, as is characteristic in Japanese passives. The B MAP is introduced by the affective union structure (as it is introduced by the causative union above), but is cancelled as part of the passive structure in Japanese. Of course, (I05/io6) is an instance of an unergative base predicate and is well-formed.

An adversative of an unaccusative would have to have a structure something like the following:

(107) 0-Rs: TH Affectee

GRs: [2]

1 2

MAPs: A 

Aside from the fact that the unlinked theme 2 would not be marked dative (unlinked themes would take accusative), the structure in (107) is ill-formed for language-particular reasons. Dubinsky (1985) convincingly shows, within an RG treatment, that all P-sectors in Japanese must contain a I in the P- final stratum (the stratum in which a particular predicate last bears the P relation). The analogue for this in Mapping Theory representations would be that all inner layers must contain a I. Note that the inner layer in (107) consists solely of a 2. This then violates the language-particular constraint on well-formed inner layers and accounts for the absence of adversatives of unaccusatives in Japanese.

7.3 A final look at Javanese

It remains to account for the unaccusative/unergative split with respect to passives in Javanese (and Indonesian). In fact, the restriction is not on passives (or adversatives) but on the distribution of the Javanese locative suffix, or in classic RG terms on advancement to 2. As we saw above,

48


unaccusatives take accidental passives (108) while unergatives seem not to (IO9).

(Io8) Montor-e Amir ke-tiba-n watu. car-DEF A AD-fall rock 'Amir's car got fallen on by a rock.'

(109) *Kertas-e Amir ke-playo-nan bocahbocah. paper-DEF A AD-run children

(Amir's paper got run on by the children.)

As outlined in section 6, the -an suffix is a reflex of the goal, location, or source argument's GR linking to a MAP, in (io8) the A MAP.

One might suppose that the ungrammaticality of (I09) can be accounted for by a stipulation such that obliques cannot link to the A MAP when the predicate is unergative. In fact, however, the restriction is broader. As the following show, locative morphology is impossible with unergatives of any kind.

( I o) Bocahbocah kuwi mlayu menyang toko kuwi. children that AF.run to store that 'The children ran to the store.'

(I I I) * Bocahbocah kuwi mlayo-ni toko kuwi. children that AF.run-LOC store that

(The children ran to the store.)

(I12) *Toko kuwi di-playo-ni bocahbocah. store that OF-run-LOC children

(The store was run to by the children.)

(I io) illustrates an unergative with a goal argument; the argument occurs in a PP marked by menyang 'to'. Linking the goal argument to a MAP yields ungrammatical results, whether it would be an added B MAP, as in (I I I), or the A MAP, as in (I I2). So, in Javanese, linking the GR of a goal, locative, or source (in fact any oblique) argument with an unergative base predicate is disallowed. Such linking is possible only with transitive and unaccusative bases, or in other words, in the presence of a 2. Thus, to the extent that there is a language-particular restriction operating,28 it is a restriction on linking oblique arguments (or in RG terms, on advancement to 2); it is not a restriction on a particular construction (i.e. adversatives).

While it is not necessary to adopt Gerdts' Mapping Theory to draw this conclusion, the Mapping Theory framework provides the apparatus for accounting for the full range of passives in Javanese (and Indonesian)

[28] One speaker I consulted in fact accepts sentences such as (i I) and (II2). However, this same speaker also accepts as fully grammatical the 'adversative' of an unergative, (109). Therefore, for this speaker, apparently the restriction on linking obliques is not operative.

49



WILLIAM D. DAVIES

without violating any principles and provides an explanation for other

troubling data as well.

REFERENCES

Bickford, J. A. (I987). Universal constraints on relationally complex clauses. Ph.D. dissertation, University of California, San Diego.

Bintoro (1980). Javanese transitive verbs: a tagmemic analysis. NUSA Linguistic Studies in Indonesian and Other Languages in Indonesia 8. Jakarta: NUSA.

Cartier, A. (1978). On Ke-verb sentences in Indonesian. In Wurm, S. A. & Carrington, L. (eds.) Second International Conference on Austronesian Linguistics: proceedings, Pacific Linguistics, Series C, No. 6I. Canberra: Australian National University.

Chung, S. (1976). An object-creating rule in Bahasa Indonesia. Linguistic Inquiry 7. 41-87. Culy, C. & Davies, W. D. (1993). Mapping Theory and Fula verbal extensions. Presented at the

6th Biennial Conference on Grammatical Relations. Vancouver, Canada. Cumming, S. (I986). World order change in Malay. In Geraghty, P., Carrington, L. & Wurm,

S. A. (eds.) FOCAL I: papers from the Fourth International Conference on Austronesian Linguistics. Canberra: Australian National University. 97-I I I.

Davies, W. D. (I984). Antipassive: Choctaw evidence for a universal characterization. In Perlmutter, D. M. & Rosen, C. (eds.) Studies in Relational Grammar 2. Chicago: University of Chicago Press. 331-376.

Davies, W. D. (1993). Javanese subjects and topics and psych verbs. Linguistics 31. 239-277. Davies, W. D. & Rosen, C. (I988). Unions as multi-predicate clauses. Language 64. 52-88. Davies, W. D. & Sam-Colop, L. (I990). K'iche' and the structure of antipassive. Language 66.

522-549. Dubinsky, S. (I985). Japanese union constructions: a unified analysis of -sase and -rare. Ph.D.

dissertation, Cornell University. Dubinsky, S. (1993). A union analysis for Japanese 'adversative passives'. Ms., University of

South Carolina. Durie, M. (I986). The use of passive-like strategies in Acehnese. In Dardjowidjojo, S. (ed.)

Miscellaneous studies in Indonesian and other languages in Indonesia, part VIII. NUSA Linguistics Studies of Indonesian and Other Languages in Indonesia. Jakarta: NUSA.

Fillmore, C. (1965). Indirect object constructions in English and the ordering of transformations. The Hague: Mouton.

Gerdts, D. B. (i988a). Antipassive and causatives in Ilokano: evidence for the ergative analysis. In McGinn, R. (ed.) Studies in Austronesian linguistics. Athens, OH: University of Ohio Press. 295-321.

Gerdts, D. B. (i988b). Object and absolutive in Halkomelem Salish. New York: Garland. Gerdts, D. B. (I990). Revaluation and inheritance in Korean causative union. In Postal, P. M.

& Joseph, B. D. (eds.) Studies in Relational Grammar 3. Chicago: University of Chicago Press. 203-246.

Gerdts, D. B. (I99I). Case spread in multipredicate domains. Chicago Linguistic Society 27. 196-218.

Gerdts, D. B. (1992). Morphologically-mediated relational profiles. Berkeley Linguistics Society I8. 322-337.

Gerdts, D. B. (I993a). Mapping Halkomelem grammatical relations. Linguistics 31. 59I-62I. Gerdts, D. B. (I993b). Mapping transitive voice in Halkomelem. Berkeley Linguistics Society

19S. 22-33. Green, G. M. (1974). Semantics and syntactic regularity. Bloomington, IN: Indiana University

Press. Home, E. C. (I96I). Beginning Javanese. New Haven: Yale University Press. Joseph, B. D. (I982). On some advancements to subject in Greek. Ohio State University

Working Papers in Linguistics 26. 49-58. Junus, U. (I97I). Ke- -an constructions in Indonesian. Linguistic 76. 24-44. Kana, M. (I986). Grammatical relations in Bahasa Indonesia. Ph.D. dissertation, Cornell

University. Kimenyi, A. (1980). A Relational Grammar of Kinyarwanda. University of California

Publications in Linguistics 91. Berkeley.

50




Legendre, G. (I990). French impersonal constructions. Natural Language and Linguistic Theory 8. 81-128.

Naylor, P. B. (1978). Toward focus in Austronesian. In Wurm, S. A. & Carrington, L. (eds.) Second International Conference on Austronesian Linguistics: proceedings. Pacific Linguistics, Series C, No. 6i. Canberra: Australian National University. 395-442.

Nerbonne, J. A. (1982). Some passives not characterized by universal rules: subjectless passives. Ohio State University Working Papers in Linguistics 26. 59-92.

Nichols, L. (I990). The so-called Javanese passive and lexically based constraints on anaphoric binding. Ms., University of Chicago.

Oehrle, R. T. (1976). The grammatical status of the English dative alternation. Ph.D. dissertation, MIT.

Ozkarag6z, I. (1986). The relational structure of Turkish syntax. Ph.D. dissertation, University of California.

Perlmutter, D. M. & Postal, P. M. (I984). The i-Advancement Exclusiveness Law. In Perlmutter, D. M. & Rosen, C. (eds.) Studies in Relational Grammar 2. Chicago: University of Chicago Press. 81-125.

Poedjosoedarmo, G. R. (I986). Role structure in Javanese. NUSA Linguistic Studies of Indonesian and Other Languages in Indonesia 24. Jakarta: NUSA.

Postal, P. M. (I977). Antipassive in French. Lingvisticae Investigationes I. 333-374. Postal, P. M. (1986). Studies of passive clauses. Albany: SUNY Press. Ramelan (I983). Javanese indicative and imperative passives. In Halim, A., Carrington, L. &

Wurm, S. A. (eds.) Papers from the Third International Conference on Austronesian Linguistics. Canberra: Australian National University. 199-214.

Richards, N. (I994). Adpositions and the English pseudo-passive. Ms., Cornell University. Smith-Hefner, N. J. (I988). Cara Tengger: notes on a non-standard dialect of Javanese. In

McGinn, R. (ed.) Studies in Austronesian linguistics. Athens, OH: University of Ohio Press. 203-233.

Suzuki, K. (I984). A relational analysis of Japanese causatives. MA dissertation, Cornell University.

Thomas, M. R. (1980). Verb affixes and focus in Bahasa Indonesia. In Naylor, P. B. (ed.) Austronesian studies: papers from the Second Eastern Conference on Austronesian Languages. Ann Arbor: University of Michigan. 63-70.

Timberlake, A. (I982). Impersonal passives in Lithuanian. Berkeley Linguistics Society 8. 508-524.

Verhaar, J. W. M. (I978). Syntactic (in)alienability in Indonesian. In Wurm, S. A. & Carrington, L. (eds.) Second International Conference on Austronesian Linguistics: proceedings, Pacific Linguistics, Series C, No. 6i. Canberra: Australian National University.

Verhaar, J. W. M. (I984). The categorial system in contemporary Indonesian: verbs. In Purwo, B. K. (ed.) Towards a description of contemporary Indonesian: preliminary studies, part I. NUSA Linguistic Studies in Indonesian and Other Languages in Indonesia. Jakarta: NUSA.

Wouk, F. (I986). Transitivity in proto-Malayo-Polynesian and proto-Austronesian. In Geraghty, P., Carrington, L. & Wurm, S. A. (eds.) FOCAL I: papers from the Fourth International Conference on Austronesian Linguistics. Canberra: Australian National Uni- versity. 133-158.

Author's address: Department of Linguistics, The University of Iowa, 570 English Philosophy Building, Iowa City, Iowa 52242-I408, U.S.A. E-mail: [email protected]

51



Javanese adversatives, passives, and Mapping Theory

Documents

Transcript of Javanese adversatives, passives, and Mapping Theory